Mixed Error Effect

The mixed error effect is a finding from the study of natural speech-error corpora in which word-substitution errors, cases where a speaker says one word in place of the intended one, are more likely to be similar to the intended word in both meaning and sound at once than would be expected if semantic similarity and sound similarity contributed to such errors independently of one another. In other words, an error such as saying rat for the intended target cat, which is similar in sound, or saying dog for cat, which is similar in meaning, occurs at roughly the rate predicted by chance for each kind of similarity alone, but an error that is similar in both respects simultaneously occurs considerably more often than that independence assumption would predict. The finding was influential in the development of interactive activation models of speech production associated with researchers including Gary Dell in the 1980s, since such models, in which activation can spread both forward from meaning toward sound and backward from partially activated sound forms toward related word meanings, naturally predict that an error candidate reinforced by both pathways at once should be especially likely to be selected, unlike models in which meaning-based and sound-based stages of production operate in strict, non-interacting sequence.